Arena Physica launched Heaviside-1, a foundation model for electromagnetism, via Show HN. Trained on 500B field samples from 250K+ physical layout designs, it predicts design fields ~10^5x faster than commercial full-wave solvers with S-parameter magnitude error under 1 dB, with a public Fields Studio demo.
Two arXiv papers propose physics-based neural operators for wireless communications. GCNO compresses channel matrices by reporting only dominant propagation paths, adapting to channel complexity and avoiding retraining when antenna counts change. A second paper introduces a physics-unrolled neural operator for radio-map prediction that handles Monte Carlo noise in training labels.
Two arXiv papers propose physics-based neural operators for wireless communications. GCNO compresses channel matrices by reporting only dominant propagation paths, adapting to channel complexity and avoiding retraining when antenna counts change. A second paper introduces a physics-unrolled neural operator for radio-map prediction that handles Monte Carlo noise in training labels.
An arXiv paper introduces a physics-guided transformer framework for transient electromigration stress analysis in multi-segment interconnect lines, addressing the rising EM reliability challenge at scaled process nodes. The model encodes lines as geometry- and DC-aware segment tokens and predicts stress far faster than matrix-exponential physics solvers.
An arXiv paper introduces a physics-guided transformer framework for transient electromigration stress analysis in multi-segment interconnect lines, addressing the rising EM reliability challenge at scaled process nodes. The model encodes lines as geometry- and DC-aware segment tokens and predicts stress far faster than matrix-exponential physics solvers.
Physicsbase.ai is a tool that enables AI agents to perform numerical simulations of physics problems using finite element analysis (FEA). It allows AI to create or submit geometries and receive validated simulation results, iterating to meet design objectives. This helps AI CAD startups integrate physics simulation into their workflows.
Physicsbase.ai is a tool that enables AI agents to perform numerical simulations of physics problems using finite element analysis (FEA). It allows AI to create or submit geometries and receive validated simulation results, iterating to meet design objectives. This helps AI CAD startups integrate physics simulation into their workflows.
A recent paper on arxiv explores how neural emulators for solving partial differential equations relate to classical numerical solvers. It argues that despite being framed as opaque alternatives, neural architectures share similarities with discretization methods, enabling spectral error analysis and potential speedups for engineering and science applications.
A recent paper on arxiv explores how neural emulators for solving partial differential equations relate to classical numerical solvers. It argues that despite being framed as opaque alternatives, neural architectures share similarities with discretization methods, enabling spectral error analysis and potential speedups for engineering and science applications.
The EMRB benchmark has been introduced to evaluate LLMs' reasoning over raw electromagnetic signals. It tests LLMs' ability to analyze I/Q data through code generation and execution. This addresses a gap in assessing LLMs for scientific and engineering tasks involving physical-layer measurements.
The EMRB benchmark has been introduced to evaluate LLMs' reasoning over raw electromagnetic signals. It tests LLMs' ability to analyze I/Q data through code generation and execution. This addresses a gap in assessing LLMs for scientific and engineering tasks involving physical-layer measurements.